overview problem saving energy (“solution”)
TRANSCRIPT
![Page 1: OVERVIEW Problem Saving Energy (“Solution”)](https://reader034.vdocuments.site/reader034/viewer/2022050211/626e279bc1717a0a2577d3e1/html5/thumbnails/1.jpg)
ENERGY-EFFICIENT DATA REPLICATION IN CLOUD COMPUTING DATACENTERS Presented by David Ocejo
![Page 2: OVERVIEW Problem Saving Energy (“Solution”)](https://reader034.vdocuments.site/reader034/viewer/2022050211/626e279bc1717a0a2577d3e1/html5/thumbnails/2.jpg)
OVERVIEW
¢ Problem
¢ Saving Energy (“Solution”) � Efficiency � Data Center Topology
¢ Simulation � Conditions � Results
![Page 3: OVERVIEW Problem Saving Energy (“Solution”)](https://reader034.vdocuments.site/reader034/viewer/2022050211/626e279bc1717a0a2577d3e1/html5/thumbnails/3.jpg)
PROBLEM
¢ Increasing energy consumption
¢ Up to 1.5% of World’s Electricity (in 2010) � from 1.0% (in 2005)
![Page 4: OVERVIEW Problem Saving Energy (“Solution”)](https://reader034.vdocuments.site/reader034/viewer/2022050211/626e279bc1717a0a2577d3e1/html5/thumbnails/4.jpg)
WORLD’S ELECTRICITY GENERATION
40%
23%
16%
11%
5% 5%
Coal
Natural Gas
Hydro
Nuclear
Oil
Other
![Page 5: OVERVIEW Problem Saving Energy (“Solution”)](https://reader034.vdocuments.site/reader034/viewer/2022050211/626e279bc1717a0a2577d3e1/html5/thumbnails/5.jpg)
DATA CENTER ENERGY CONSUMPTION
45%
15%
40%
Cooling
Power Distribution
Networking
Servers
![Page 6: OVERVIEW Problem Saving Energy (“Solution”)](https://reader034.vdocuments.site/reader034/viewer/2022050211/626e279bc1717a0a2577d3e1/html5/thumbnails/6.jpg)
ENERGY EFFICIENCY
¢ Two approaches:
� Shutting down components
� Scaling down performance
![Page 7: OVERVIEW Problem Saving Energy (“Solution”)](https://reader034.vdocuments.site/reader034/viewer/2022050211/626e279bc1717a0a2577d3e1/html5/thumbnails/7.jpg)
ENERGY EFFICIENCY
¢ Shutting Down Components
� Dynamic Power Management (DPM)
� Dynamic Network Shutdown (DNS)
![Page 8: OVERVIEW Problem Saving Energy (“Solution”)](https://reader034.vdocuments.site/reader034/viewer/2022050211/626e279bc1717a0a2577d3e1/html5/thumbnails/8.jpg)
ENERGY EFFICIENCY
¢ Scaling Down Performance
� Dynamic Voltage and Frequency Scaling (DVFS) ¢ Applicable only to CPU ¢ Other components still consume at peak rates
� Dynamic Voltage Scaling (DVS) ¢ Links
� P = V2 * f ¢ = (supplied voltage 2) * (operating frequency)
![Page 9: OVERVIEW Problem Saving Energy (“Solution”)](https://reader034.vdocuments.site/reader034/viewer/2022050211/626e279bc1717a0a2577d3e1/html5/thumbnails/9.jpg)
ENERGY EFFICIENCY
¢ Virtualization
![Page 10: OVERVIEW Problem Saving Energy (“Solution”)](https://reader034.vdocuments.site/reader034/viewer/2022050211/626e279bc1717a0a2577d3e1/html5/thumbnails/10.jpg)
OUR DATA REPLICATION APPROACH
¢ Joint optimization of energy consumption and bandwidth capacity
¢ Optimization of communication delays
![Page 11: OVERVIEW Problem Saving Energy (“Solution”)](https://reader034.vdocuments.site/reader034/viewer/2022050211/626e279bc1717a0a2577d3e1/html5/thumbnails/11.jpg)
DATA CENTER
¢ Three Tier Topology
� Core Layer ¢ Flows going in and out of data center
� Aggregation Layer ¢ Integrates connections and traffic flows from racks
� Access Layer ¢ Where computing servers are arranged into racks
![Page 12: OVERVIEW Problem Saving Energy (“Solution”)](https://reader034.vdocuments.site/reader034/viewer/2022050211/626e279bc1717a0a2577d3e1/html5/thumbnails/12.jpg)
DATA CENTER
![Page 13: OVERVIEW Problem Saving Energy (“Solution”)](https://reader034.vdocuments.site/reader034/viewer/2022050211/626e279bc1717a0a2577d3e1/html5/thumbnails/13.jpg)
DATA CENTER
![Page 14: OVERVIEW Problem Saving Energy (“Solution”)](https://reader034.vdocuments.site/reader034/viewer/2022050211/626e279bc1717a0a2577d3e1/html5/thumbnails/14.jpg)
DATA CENTER
¢ External requests directed to Rack DB � If necessary, Database DB and Central DB
¢ Databases maintain and exchange access records � Requesting (rack) server and database � Number of data item accesses and updates
¢ Popularity � Access rate: number of access events in given time
period � Decays
![Page 15: OVERVIEW Problem Saving Energy (“Solution”)](https://reader034.vdocuments.site/reader034/viewer/2022050211/626e279bc1717a0a2577d3e1/html5/thumbnails/15.jpg)
DATA CENTER
![Page 16: OVERVIEW Problem Saving Energy (“Solution”)](https://reader034.vdocuments.site/reader034/viewer/2022050211/626e279bc1717a0a2577d3e1/html5/thumbnails/16.jpg)
DATA CENTER TRANSMISSIONS
¢ Uplink – Bandwidth � Propagating database requests � Updating data items
¢ Downlink – Bandwidth � Delivering workload descriptions � Receiving database objects � Propagating updates between DB replicas
![Page 17: OVERVIEW Problem Saving Energy (“Solution”)](https://reader034.vdocuments.site/reader034/viewer/2022050211/626e279bc1717a0a2577d3e1/html5/thumbnails/17.jpg)
POWER CONSUMPTION - SERVERS
¢ Servers consume two-thirds when idle � Memory modules, disks, I/O, etc. still consuming at
peak rate
)1)(2
(1aeloadFixedPeakFixed
−−+
−+=
![Page 18: OVERVIEW Problem Saving Energy (“Solution”)](https://reader034.vdocuments.site/reader034/viewer/2022050211/626e279bc1717a0a2577d3e1/html5/thumbnails/18.jpg)
POWER CONSUMPTION - SWITCHES
� Power drawn by port running at rate r � Number of ports running at rate r � Utilization of ports
¢ 85-97% fixed energy consumption ¢ 3-15% consumed by port transceivers
∑=
++=R
r
rp
rp
rp uPnLineCardneCardsNumberOfLiChassis
1)**()*(
![Page 19: OVERVIEW Problem Saving Energy (“Solution”)](https://reader034.vdocuments.site/reader034/viewer/2022050211/626e279bc1717a0a2577d3e1/html5/thumbnails/19.jpg)
SIMULATION
¢ Performed using GreenCloud simulator � Cloud computing simulator � Packet level communication
¢ Single data center simulation � 60 minutes
![Page 20: OVERVIEW Problem Saving Energy (“Solution”)](https://reader034.vdocuments.site/reader034/viewer/2022050211/626e279bc1717a0a2577d3e1/html5/thumbnails/20.jpg)
SIMULATION – CONDITIONS
![Page 21: OVERVIEW Problem Saving Energy (“Solution”)](https://reader034.vdocuments.site/reader034/viewer/2022050211/626e279bc1717a0a2577d3e1/html5/thumbnails/21.jpg)
SIMULATION – CONDITIONS
¢ DB queries limited to 1500 bytes � Fits into single Ethernet packet
¢ Varying: � Data item size � Data access and update rates � Replication threshold
¢ DNS power saving enabled
![Page 22: OVERVIEW Problem Saving Energy (“Solution”)](https://reader034.vdocuments.site/reader034/viewer/2022050211/626e279bc1717a0a2577d3e1/html5/thumbnails/22.jpg)
SIMULATION – RESULTS
![Page 23: OVERVIEW Problem Saving Energy (“Solution”)](https://reader034.vdocuments.site/reader034/viewer/2022050211/626e279bc1717a0a2577d3e1/html5/thumbnails/23.jpg)
SIMULATION – RESULTS
![Page 24: OVERVIEW Problem Saving Energy (“Solution”)](https://reader034.vdocuments.site/reader034/viewer/2022050211/626e279bc1717a0a2577d3e1/html5/thumbnails/24.jpg)
SIMULATION – RESULTS
![Page 25: OVERVIEW Problem Saving Energy (“Solution”)](https://reader034.vdocuments.site/reader034/viewer/2022050211/626e279bc1717a0a2577d3e1/html5/thumbnails/25.jpg)
SIMULATION – RESULTS
![Page 26: OVERVIEW Problem Saving Energy (“Solution”)](https://reader034.vdocuments.site/reader034/viewer/2022050211/626e279bc1717a0a2577d3e1/html5/thumbnails/26.jpg)
CONCLUSION
¢ Replicating data closer to data consumers reduces:
� Energy consumption
� Bandwidth usage
� Communication delays
¢ Degree of reduction dependant on update rate